Abstract:•In this paper, a novel graph learning method is proposed to learn a more interpretable and robust graph for data clustering.•The global and local structures are simultaneously exploited for graph learning, which ensures to learn a more reasonable graph.•The learned graph has good interpretability to samples by introducing the nonnegative constraint to the model.•The proposed method is robust to noise and redundant features by introducing an adaptive weighted matrix.

Adaptive Weighted Nonnegative Low-Rank Representation.